Can we teach machines to develop medicine?
Posted on February 29, 2016
Today, we know molecular basis for almost over 4000 diseases. On the other hand, we have the treatment for only 250 of them. One would think that once you got the mechanism of the disease right, developing of some kind of treatment would be easy, you would think that once you have the knowledge, it’s just a matter of time when you will find a cure. Well, it turns out it is a matter of time, a very long time.
Why development of new drugs takes such a long time? What is a drug in a first place? Drugs, are usually made of a small molecules containing hydrogen, nitrogen, oxygen, carbon and few other atoms forming together a shape. That shape of the molecule is quite important as it decides if that particular molecule will find it’s target and act on it, or it will pass a long and finally be expelled from the body. So on one side we have a molecular basis of a disease (usually consisted of one or more malfunctioning large molecules like proteins) and on the other, a small molecule capable of interacting with a large molecule and stop it’s malfunctioning.
How does the current drug development look like? First step of the process is finding the target, once you figured out the mechanism of the disease, you choose the best point of that mechanism in which your drug will then go to interfere. Once the target is chosen you have to find a way to find out which type of molecules could be a hit. Triage of your hits will tell you which shapes are good candidates and which are to avoid. you then go to optimize your lead drug. Here you enter the realm of a classic medicinal chemistry, where you change some parts of your molecule in order to improve its properties like potency, selectivity, toxicity and so on. Once you have narrowed down your choice to two or three candidates you enter something that is called pre-clinical development where you have to give answers whether or not your lead compounds can be produced in big scales, you have to find out will your synthesis be reproducible at larger scales, what kind of yield will your compound give and finally how pure can you make it to be. Ultimately, you have to prove that your compound is not toxic, and then you can engage the clinical trials which is the most expensive and the longest part. Knowing all of this it is not so strange that typically the development of a new drug lasts 14 years.
We in Heuro Labs have asked ourselves. Can we do better? Can we make it go faster? One way to go faster is to take advantage of new technologies.
Not so long ago in order to make the computer do some particular task, you would have to program it. You would have to write a code in excruciating detail every single step of the way in order to achieve your goal. If you want that computer does something which you don’t know how to do yourself, that would be very difficult problem to solve.
Luckily, today this is not a case. Today we have computers which can beat the best human chessmasteres, they can drive cars in batter/safer manner than any human driver, they can even listen and describe pictures, but can computers help us to make better drugs? Can the computers help us to speed up the drug developing process?
As mentioned above the longest and most expensive phase of development is the clinical phase. There are different use cases for artificial intelligence to support in optimizing the clinical trial phase. This include faster interpretation of the data, predicting outcome and also expanding the coverage without incurring additional major costs. Additionally, we could get some major assistance from the computers on the initial phase of drug development. Moreover, we could screen all of already synthesized drugs which already passed some kind of clinical trials or at least they are considered safe for human use and test them as potential lead drugs in some other condition different from the one they were initially developed.
Today computers are already largely used in drug development. Frequently, before engaging in any synthesis molecules are evaluated in silico in order to figure out how are they fitting their protein targets. However, this process requires knowledge of the protein structure and long calculation times which even on our best computer clusters last for days or even weeks. We at Heuro Labs believe that this process can be made simpler.
However, the search for the lead drug and its optimisation is not an only point where we see possibility of improving our drug development process. Computers could help us also in synthesis department where we could get fast answer if the synthesis of desired molecule is feasible. We could receive a hint of the best path how to make that particular molecule and how easy or difficult scaling up of the whole process will be.
We think that are current drug development is flawed and as such represents major possibilities for improvement. We believe that development of machine learning algorithms could give a major boost to the field.
Want to receive updates on our progress? Signup for our newsletter